Currently, the compiler already generates common symbols for type
descriptors, so the type descriptors are unique. However, when a
type is created through reflection, it is not deduplicated with
compiler-generated types. As a consequence, we cannot assume type
descriptors are unique, and cannot use pointer equality to
compare them. Also, when constructing a reflect.Type, it has to
go through a canonicalization map, which introduces overhead to
reflect.TypeOf, and lock contentions in concurrent programs.
In order for the reflect package to deduplicate types with
compiler-created types, we register all the compiler-created type
descriptors at startup time. The reflect package, when it needs
to create a type, looks up the registry of compiler-created types
before creates a new one. There is no lock contention since the
registry is read-only after initialization.
This lets us get rid of the canonicalization map, and also makes
it possible to compare type descriptors with pointer equality.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179598
From-SVN: r271894
When the runtime collects a stack trace to associate it with some
profiling event (mem alloc, mutex, etc) there is a skip count passed
to runtime.Callers (or equivalent) to skip some known count of frames
in order to get to the "interesting" frame corresponding to the
profile event. Now that the profiling mechanism uses lazy fixup (when
removing compiler artifacts like thunks, morestack calls etc), we also
need to move the frame skipping logic after the fixup, so as to insure
that the skip count isn't thrown off by these artifacts.
Fixesgolang/go#32290.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179740
From-SVN: r271892
Revise the gccgo version of memory/block/mutex profiling to reduce
runtime overhead. The main change is to collect raw stack traces while
the profile is on line, then post-process the stacks just prior to the
point where we are ready to use the final product. Memory profiling
(at a very low sampling rate) is enabled by default, and the overhead
of the symbolization / DWARF-reading from backtrace_full was slowing
things down relative to the main Go runtime.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/171497
From-SVN: r271172
runtime.throw needs a g to work properly. Set up g early, to
ensure that if something goes wrong in the runtime startup (e.g.
runtime.check fails), the program terminates in a reasonable way.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/176657
From-SVN: r271088
A direct interface is an interface whose data word contains the
actual data value, instead of a pointer to it. The gc toolchain
creates a direct interface if the value is pointer shaped, that
includes pointers (including unsafe.Pointer), functions, channels,
maps, and structs and arrays containing a single pointer-shaped
field. In gccgo, we only do this for pointers. This CL unifies
direct interface types with gc. This reduces allocations when
converting such types to interfaces.
Our method functions used to always take pointer receivers, to
make interface calls easy. Now for direct interface types, their
value methods will take value receivers. For a pointer to those
types, when converted to interface, the interface data contains
the pointer. For that interface to call a value method, it will
need a wrapper method that dereference the pointer and invokes
the value method. The wrapper method, instead of the actual one,
is put into the itable of the pointer type.
In the runtime, adjust funcPC for the new layout of interfaces of
functions.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/168409
From-SVN: r270779
Previously, each time we do an interface conversion for which the
method table is not known at compile time, we allocate a new
method table.
This CL ports the mechanism of itab caching from the gc runtime,
adapted to our itab representation and method finding mechanism.
With the cache, we reuse the same itab for the same (interface,
concrete) type pair. This reduces allocations in interface
conversions.
Unlike the gc runtime, we don't prepopulate the cache with
statically allocated itabs, as currently we don't have a way to
find them. This means we don't deduplicate run-time allocated
itabs with compile-time allocated ones. But that is not too bad
-- it is just a cache anyway.
As now itabs are never freed, it is also possible to drop the
write barrier for writing the first word of an interface header.
I'll leave this optimization for the future.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/171617
From-SVN: r270778
AIX doesn't allow to mmap an address range which is already mmap.
Therefore, once the region has been allocated, it must munmap before
being able to play with it.
The corresponding Go Toolchain patch is CL 174059.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/174138
From-SVN: r270615
In the C calling convention, on AMD64, and probably a number of
other architectures, a 3-word struct argument is passed on stack.
This is less efficient than passing in three registers. Further,
this may affect the code generation in other part of the program,
even if the function is not actually called.
Slices are common in Go and append is a common slice operation,
which calls growslice in the growing path. To improve the code
generation, pass the slice header's three fields as separate
values, instead of a struct, to growslice.
The drawback is that this makes the runtime implementation
slightly diverges from the gc runtime.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/168277
From-SVN: r269811
Since aix/ppc64 has been added to GC toolchain, a mix between new and
old files were created in gcc toolchain.
This commit corrects this merge for aix/ppc64 and aix/ppc.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/167658
From-SVN: r269797
PR go/89447
syscall, internal/syscall: adjust use of largefile functions
Consistently call __go_openat for openat. Use fstatat64, creat64,
sendfile64, and getdents64 where needed.
Based on patch by Rainer Orth.
Fixes https://gcc.gnu.org/PR89447
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/166420
From-SVN: r269521
In the runtime there are bad pointer checks that currently don't
work with the concervative collector. With stack maps, the GC is
precise and the checks should work. Enable them.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/153871
From-SVN: r269406
Interpreting auxv as []uintptr is incorrect on 64-bit big-endian,
as auxv alternates a 32-bit int with a 64-bit pointer.
Patch from Rainer Orth.
Reviewed-on: https://go-review.googlesource.com/c/164739
From-SVN: r269315
When using the go tool with gccgo, this changes the default
compilation to use -O2. The -gccgoflags option can be used to
override this default. I think this change better corresponds to what
people expect when using the go tool.
Reviewed-on: https://go-review.googlesource.com/c/164378
From-SVN: r269299
PR go/89172
internal/cpu, runtime, runtime/pprof: handle function descriptors
When using PPC64 ELF ABI v1 a function address is not a PC, but is the
address of a function descriptor. The first field in the function
descriptor is the actual PC (see
http://refspecs.linuxfoundation.org/ELF/ppc64/PPC-elf64abi.html#FUNC-DES).
The libbacktrace library knows about this, and libgo uses actual PC
values consistently except for the helper function funcPC that appears
in both runtime and runtime/pprof.
This patch fixes funcPC by recording, in the internal/cpu package,
whether function descriptors are being used. We have to check for
function descriptors using a C compiler check, because GCC can be
configured using --with-abi to select the ELF ABI to use.
Fixes https://gcc.gnu.org/PR89172
Reviewed-on: https://go-review.googlesource.com/c/162978
From-SVN: r269266
Backport https://golang.org/cl/163237 from the master library:
Ensure that cmd/go consistently calls base.Exit rather than os.Exit,
so that we don't incorrectly leave the work directory around on exit.
Test this by modifying the testsuite to run all the tests with TMPDIR
set to a temporary directory, and then check that no files are left
behind in that temporary directory. Adjust a couple of tests to make
this approach work.
Updates https://gcc.gnu.org/PR89406
Reviewed-on: https://go-review.googlesource.com/c/163198
From-SVN: r269086
In signal-triggered stack scan, if the signal is delivered at
certain bad time (e.g. in vdso, or in the middle of setcontext?),
the unwinder may not be able to unwind the whole stack, while it
still reports _URC_END_OF_STACK. So we cannot rely on _URC_END_OF_STACK
to tell if it successfully scanned the stack. Instead, we check
the last Go frame to see it actually reached the end of the stack.
For Go-created stack, this is runtime.kickoff. For C-created
stack, we need to record the outermost Go frame when it enters
the Go side.
Also we cannot unwind the stack if the signal is delivered in the
middle of runtime.gogo, halfway through a goroutine switch, where
the g and the stack don't match. Give up in this case as well.
Reviewed-on: https://go-review.googlesource.com/c/159098
From-SVN: r269018
Compiling with LTO revealed a number of cases in the runtime and
standard library where C and Go disagreed about the type of an object or
function (or where Go and code generated by the compiler disagreed). In
all cases the underlying representation was the same (e.g., uintptr vs.
void*), so this wasn't causing actual problems, but it did result in a
number of annoying warnings when compiling with LTO.
Reviewed-on: https://go-review.googlesource.com/c/160700
From-SVN: r268923
PR go/89168
libgo: change gotest to run examples with output
Change the gotest script to act like "go test" and run examples that
have "output" comments. This is not done with full generality, but
just enough to run the libgo tests. Other packages should be tested
with "go test" as usual.
While we're here clean up some old bits of gotest, and only run
TestXXX functions that are actually in *_test.go files. The latter
change should fix https://gcc.gnu.org/PR89168.
Reviewed-on: https://go-review.googlesource.com/c/162139
From-SVN: r268922
PR go/89199
sync/atomic: use strong form of atomic_compare_exchange_n
In the recent change to use atomic_compare_exchange_n I thought we
could use the weak form, which can spuriously fail. But that is not
how it is implemented in the gc library, and it is not what the rest
of the library expects.
Thanks to Lynn Boger for identifying the problem.
Fixes https://gcc.gnu.org/PR89199
Reviewed-on: https://go-review.googlesource.com/c/161359
From-SVN: r268591
GCC has supported the __atomic intrinsics since 4.7. They are better
than the __sync intrinsics in that they specify a memory model and,
more importantly for our purposes, they are reliably implemented
either in the compiler or in libatomic.
Fixes https://gcc.gnu.org/PR52084
Reviewed-on: https://go-review.googlesource.com/c/160820
From-SVN: r268458
If sigtramp and sigtrampgo are both on stack, n -= framesToDiscard
is executed twice, which should actually run only once.
Reviewed-on: https://go-review.googlesource.com/c/159238
From-SVN: r268366
If a panic happens in the runtime we turn that into a fatal error.
We use the caller's PC to determine if the panic call is inside
the runtime. getcallerpc returns the PC immediately after the
call instruction. If the call is the very last instruction of a
function, it may not find this PC belong to a runtime function,
giving false result. We need to back off the PC by 1 to the call
instruction.
The gc runtime doesn't do this because the gc compiler always
emit an instruction following a panic call, presumably an UNDEF
instruction which turns into an architecture-specific illegal
instruction. Our compiler doesn't do this.
Reviewed-on: https://go-review.googlesource.com/c/159437
From-SVN: r268358
Precise stack scan uses SIGURG to trigger a stack scan. We need
to have Go signal handler installed for SIGURG.
Reviewed-on: https://go-review.googlesource.com/c/159097
From-SVN: r268230
For the gofrontend copy, change calls to types.SizesFor to pass
"gccgo" rather than "gc". Leave the asmdecl pass unchanged since that
pass is gc-specific anyhow.
This has been fixed in a better way in the external repo by
https://golang.org/cl/158317 and friends, but that is not in 1.12, so
use this approach for now.
Reviewed-on: https://go-review.googlesource.com/c/158842
From-SVN: r268153
PR go/88927
runtime, internal/cpu: fix build for ARM GNU/Linux
Was failing with
../../../libgo/go/internal/cpu/cpu.go:138:2: error: reference to undefined name 'doinit'
138 | doinit()
| ^
Fix it by adding in Go 1.12 internal/cpu/cpu_arm.go, and the code in
runtime that initializes the values.
Fixes https://gcc.gnu.org/PR88927.
Reviewed-on: https://go-review.googlesource.com/c/158717
From-SVN: r268131
Restore some of the fixes that were applied to golang_org/x/net/lif
but were lost when 1.12 moved the directory to internal/x/net/lif.
Add support for reading /proc to fetch argc/argv/env for c-archive mode.
Reviewed-on: https://go-review.googlesource.com/c/158640
From-SVN: r268130
Reviewed-on: https://go-review.googlesource.com/c/158019
gotools/:
* Makefile.am (go_cmd_vet_files): Update for Go1.12beta2 release.
(GOTOOLS_TEST_TIMEOUT): Increase to 600.
(check-runtime): Export LD_LIBRARY_PATH before computing GOARCH
and GOOS.
(check-vet): Copy golang.org/x/tools into check-vet-dir.
* Makefile.in: Regenerate.
gcc/testsuite/:
* go.go-torture/execute/names-1.go: Stop using debug/xcoff, which
is no longer externally visible.
From-SVN: r268084
Currently, we dropg (which clears gp.m) after we CAS the g status
to _Grunnable or _Gwaiting. Immediately after CASing the g status,
another thread may CAS it to _Gscan status and scan its stack.
With precise stack scan, it accesses gp.m in order to switch to g
and back (in doscanstackswitch). This races with dropg. If
doscanstackswitch reads gp.m, then dropg runs, when we restore
the m at the end of the scan it will set to a stale value. Worse,
if dropg runs after doscanstackswitch sets the new m, gp will be
running with a nil m.
To fix this, we do dropg before CAS g status to _Grunnable or
_Gwaiting. We can do this safely if we are CASing from _Grunning,
as we own the g when it is in _Grunning. There is one case where
we CAS from _Gsyscall to _Grunnable. It is not safe to dropg when
it is in _Gsyscall, as precise stack scan needs to read gp.m in
order to signal the m. So we need to introduce a transient state,
_Gexitingsyscall, between _Gsyscall and _Grunnable, where the GC
should not scan its stack.
In is a little unfortunate that we have to add another g status.
We could reuse an existing one (e.g. _Gcopystack), but it is
clearer and safer to just use a new one, as Austin suggested.
Reviewed-on: https://go-review.googlesource.com/c/158157
From-SVN: r268001
Many C syscall functions take pointer arguments. The pointers
don't escape in the C functions. Mark the C functions noescape so
calling them doesn't need allocation.
Reviewed-on: https://go-review.googlesource.com/c/158158
From-SVN: r267989
CL 157557 changes the compiler to add one byte padding to
non-empty struct ending with a zero-sized field. Add the same
padding to the FFI type, so reflect.Call works.
This fixes test/fixedbugs/issue26335.go in the main repo.
Reviewed-on: https://go-review.googlesource.com/c/158018
From-SVN: r267956
This ports https://golang.org/cl/155918 from the master repo.
runtime: panic on uncomparable map key, even if map is empty
Reorg map flags a bit so we don't need any extra space for the extra flag.
This is a pre-req for updating libgo to the Go 1.12beta2 release.
Reviewed-on: https://go-review.googlesource.com/c/157858
From-SVN: r267950
This is the gccgo version of https://golang.org/cl/141822:
Only return a pointer p to the new slices backing array from makeslice.
Makeslice callers then construct sliceheader{p, len, cap} explictly
instead of makeslice returning the slice.
This change caused the GCC backend to break the runtime/pprof test by
merging together the identical functions allocateReflectTransient and
allocateTransient2M. This caused the traceback to be other than
expected. Fix that by making the functions not identical.
This is a step toward updating libgo to the Go1.12beta1 release.
Reviewed-on: https://go-review.googlesource.com/c/155937
From-SVN: r267660
Currently, when collecting a traceback for another goroutine,
getTraceback calls gogo(gp) switching to gp, which will resume in
mcall, which will call gtraceback, which will set up gp->m. There
is a gap between setting the current running g to gp and setting
gp->m. If a profiling signal arrives in between, sigtramp will
see a non-nil gp with a nil m, and will seg fault. Fix this by
setting up gp->m first.
Fixesgolang/go#29448.
Reviewed-on: https://go-review.googlesource.com/c/156038
From-SVN: r267658
This increases the time to wait for signals to be delivered in the
TestAtomicStop testcase. When running gccgo tests on ppc64 or ppc64le,
there are intermittent failures in this test because the wait time is
too small.
Updates golang/go#29046
Reviewed-on: https://go-review.googlesource.com/c/153879
From-SVN: r267068
This CL adds support of precise stack scan using stack maps to
the runtime. The stack maps are generated by the compiler (if
supported). Each safepoint is associated with a (real or dummy)
landing pad, and its "type info" in the exception table is a
pointer to the stack map. When a stack is scanned, the stack map
is found by the stack unwinding code by inspecting the exception
table (LSDA).
For precise stack scan we need to unwind the stack. There are
three cases:
- If a goroutine is scanning its own stack, it can unwind the
stack and scan the frames.
- If a goroutine is scanning another, stopped, goroutine, it
cannot directly unwind the target stack. We handle this by
switching (runtime.gogo) to the target g, letting it unwind
and scan the stack, and switch back.
- If we are scanning a goroutine that is blocked in a syscall,
we send a signal to the target goroutine's thread, and let the
signal handler unwind and scan the stack. Extra care is needed
as this races with enter/exit syscall.
Currently this is only implemented on linux.
Reviewed-on: https://go-review.googlesource.com/c/140518
From-SVN: r266832
In buildmodeinit, the c-archive buildmode is flagged as invalid
on linux/ppc64 for gccgo when it should be valid. This happens
because the check against the gccgo flag is done after the checks
for valid GOOS/GOARCH pairs instead of before as is done for all
other buildmode cases in this switch. This corrects the problem and
allows several of the gccgo gotools testcases to pass on linux/ppc64.
Updates #29046
Reviewed-on: https://go-review.googlesource.com/c/152137
From-SVN: r266764
For inlined function bodies we're going to need to refer to variables,
so change the existing export data to add a '$' to names that look
like identifiers: true, false, nil, convert.
While we're here drop an unnecessary space character after operators.
Reviewed-on: https://go-review.googlesource.com/c/150067
From-SVN: r266529
Create a framework for putting function bodies in export data. At
present only empty functions will be put there, and they will be
ignored on import. Later patches will get this to the point of
supporting inlining of (some) functions defined in other packages.
Reviewed-on: https://go-review.googlesource.com/c/150061
From-SVN: r266490
The code to implement new-style gccgo name mangling had a recipe that
didn't quite match the one in the compiler (incorrect handling for
'.'). This showed up as a failure in the gotools cgo test if the
directory containing the test run included a "." character.
Reviewed-on: https://go-review.googlesource.com/c/147917
From-SVN: r265981
Fix asm name directive for the C version of log/syslog.syslog_c,
which didn't get included in the recent name mangling change.
Reviewed-on: https://go-review.googlesource.com/c/145017
From-SVN: r265533
The current implementation of Gogo::pkgpath_for_symbol was written in
a way that allowed two distinct package paths to map to the same
symbol, which could cause collisions at link- time or compile-time.
Switch to a better mangling scheme to insure that we get a unique
packagepath symbol for each package. In the new scheme instead of having
separate mangling schemes for identifiers and package paths, the
main identifier mangler ("go_encode_id") now handles mangling of
both packagepath characters and identifier characters.
The new mangling scheme is more intrusive: "foo/bar.Baz" is mangled as
"foo..z2fbar.Baz" instead of "foo_bar.Baz". To mitigate this, this
patch also adds a demangling capability so that function names
returned from runtime.CallersFrames are converted back to their
original unmangled form.
Changing the pkgpath_for_symbol scheme requires updating a number of
//go:linkname directives and C "__asm__" directives to match the new
scheme, as well as updating the 'gotest' driver (which makes
assumptions about the correct mapping from pkgpath symbol to package
name).
Fixesgolang/go#27534.
Reviewed-on: https://go-review.googlesource.com/c/135455
From-SVN: r265510
PR go/87661
runtime: remove unused armArch, hwcap and hardDiv
After CL 140057 these are only written but never read in gccgo.
Reviewed-on: https://go-review.googlesource.com/c/141077
From-SVN: r265439
Introduce a new "types" command to the export data to record the
number of types and the size of their export data. It is immediately
followed by new "type" commands that can be indexed. Parse all the
exported types immediately so that we register them, but parse other
type data only as needed.
Reviewed-on: https://go-review.googlesource.com/c/143022
From-SVN: r265409
Previously when export data referred to a type that was not defined in
a directly imported package, we would write the package name as
additional information in the type's export data. That approach
required all type information to be read in order. This patch changes
the compiler to find all references to indirectly imported packages,
and write them out as an indirectimport line in the import data. This
will permit us to read exported type data out of order.
The type traversal used to find indirect imports is a little more
complicated than necessary in preparation for later patches in this
series.
Reviewed-on: https://go-review.googlesource.com/c/143020
From-SVN: r265296
The export data, which is approximately readable and looks something
like Go, was first implemented back when Go still used semicolons.
Drop the semicolons, to make it look slightly more Go like and make it
slightly smaller.
This updates the compiler and the gccgoimporter package.
This introduces a new version of the export data. There are going to
be more changes to the export data, so this version is still subject
to change.
Reviewed-on: https://go-review.googlesource.com/c/143018
From-SVN: r265284
LLVM doesn't support non-call exception. This test was passing
more or less by luck: if the faulting instruction is between two
calls with the same landing pad (in instruction layout order,
not the program's logic order), it generates a merged PC range
that covers the faulting instruction. If the instruction layout
order changes, or it uses two different (but may be degenerate)
landing pads, this doesn't work.
Reviewed-on: https://go-review.googlesource.com/c/140517
From-SVN: r264985
Use inline assembly in the implementation of internal_cpu.xgetbv as
opposed to a call to the intrinsic _xgetbv(), since non-gcc compilers
(e.g. clang) may or may not have support for it.
Reviewed-on: https://go-review.googlesource.com/c/140137
From-SVN: r264882
This reportedly happens on CentOS 5.11. The real code will work fine;
this test is assuming that the unexported slice function will handle
the splice, but if pipe2 does not work then it doesn't. The relevant
code in internal/poll/splice_linux.go says "Falling back to pipe is
possible, but prior to 2.6.29 splice returns -EAGAIN instead of 0 when
the connection is closed."
Reviewed-on: https://go-review.googlesource.com/138838
From-SVN: r264793
This is enough to let libgo build when configured using
--with-multilib-list=m64,m32,mx32. I don't have an x32-enabled kernel
so I haven't tested whether it executes correctly.
For https://gcc.gnu.org/PR87470
Reviewed-on: https://go-review.googlesource.com/138817
From-SVN: r264772
On Alpha GNU/Linux there is no geteuid system call, there is only
getresuid. The raw geteuid system call is only used for testing, so
just skip the test if it's not available.
Reviewed-on: https://go-review.googlesource.com/137655
From-SVN: r264647
In internal/bytealg correct a +build tag to never build indexbyte_generic.go
for the gofrontend, where we always use indexbyte_native.go.
For internal/cpu let the Makefile define CacheLineSize using goarch.sh,
rather than trying to enumerate all the possibilities in cpu_ARCH.go files.
In internal/poll call the C fcntl function rather than using SYS_FCNTL.
Change mksysinfo.sh to ensure that F_GETPIPE_SZ is always defined,
and check that in internal/poll.
Reviewed-on: https://go-review.googlesource.com/137256
From-SVN: r264572
This permits TestScript to work when gccgo is not installed.
Previous testing was using a previously installed gccgo, not the newly
built one.
This revealed that the testing of whether an internal package is
permitted was incorrect for standard library packages, since the
uninstalled gccgo can see internal packages in the uninstalled libgo.
Fix the internal package tests.
This permitted removing a couple of gccgo-specific changes in the
testsuite.
Reviewed-on: https://go-review.googlesource.com/137255
From-SVN: r264570
Reviewed-on: https://go-review.googlesource.com/136435
gotools/:
* Makefile.am (mostlyclean-local): Run chmod on check-go-dir to
make sure it is writable.
(check-go-tools): Likewise.
(check-vet): Copy internal/objabi to check-vet-dir.
* Makefile.in: Rebuild.
From-SVN: r264546
In 1.11 writebarrierptr is going away, so change the compiler to call
gcWriteBarrier instead. We weren't using gcWriteBarrier before;
adjust the implementation to use the putFast method.
This revealed a problem in the kickoff function. When using cgo,
kickoff can be called on the g0 of an m allocated by newExtraM. In
that case the m will generally have a p, but systemstack may be called
by wbBufFlush as part of flushing the write barrier buffer. At that
point the buffer is full, so we can not do a write barrier. So adjust
the existing code in kickoff so that in the case where we are g0,
don't do any write barrier at all.
Reviewed-on: https://go-review.googlesource.com/131395
From-SVN: r264295
In the sweep code we can sometimes see incorrect counts when
conservative stack scanning causes us to grey an object that we
earlier decided could be freed. We already ignored this check, but
adjust this case to maintain correct span counts when it happens.
This gives us slightly more correct numbers in MemStats, and helps
avoid a rare failure in TestReadMemStats.
Also fix the free index, and cope with finding a full span when
allocating a new one.
Reviewed-on: https://go-review.googlesource.com/134216
From-SVN: r264294
Unlike the gc runtime, libgo stores traceback information in location
structs, which contain strings. Therefore, copying location structs
around appears to require write barriers, although in fact write
barriers are never important because the strings are never allocated
in Go memory. They come from libbacktrace.
Some of the generated write barriers come at times when write barriers
are not permitted. For example, exitsyscall, marked
nowritebarrierrec, calls exitsyscallfast which calls traceGoSysExit
which calls traceEvent which calls traceStackID which calls
trace.stackTab.put which copies location values into memory allocated
by tab.newStack. This write barrier can be invoked when there is no
p, causing a crash.
This change fixes the problem by ensuring that location values are
copied around in the tracing code with no write barriers.
This was found by fixing the compiler to fully implement
//go:nowritebarrierrec; CL to follow.
Reviewed-on: https://go-review.googlesource.com/134226
From-SVN: r264282
To reduce the amount of time spent in write barrier processing
(specifically runtime.bulkBarrierPreWrite), add support for building a
'GC roots index', basically a sorted list of all roots, so as to
allow more efficient lookups of gcdata structures for globals. The
previous implementation worked on the raw (unsorted) roots list
itself, which did not scale well.
Reviewed-on: https://go-review.googlesource.com/132595
From-SVN: r264276
